Performance Modeling and Benchmarking of Event-Based ... - DVS
Performance Modeling and Benchmarking of Event-Based ... - DVS
Performance Modeling and Benchmarking of Event-Based ... - DVS
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
92 CHAPTER 5. BENCHMARKING OF EVENT-BASED SYSTEMS<br />
No. Msg.<br />
9000<br />
8000<br />
7000<br />
6000<br />
5000<br />
4000<br />
3000<br />
2000<br />
1000<br />
0<br />
40 80 120 160 200 240 280<br />
Base<br />
a<br />
b<br />
c<br />
d<br />
Kbytes<br />
20000<br />
18000<br />
16000<br />
14000<br />
12000<br />
10000<br />
8000<br />
6000<br />
4000<br />
2000<br />
0<br />
40 80 120 160 200 240 280<br />
Base<br />
a<br />
b<br />
c<br />
d<br />
Figure 5.9: Horizontal Topology: # msg.<br />
sent<br />
Figure 5.10: Message Traffic in Kbytes<br />
type <strong>and</strong> the b<strong>and</strong>width they use are scaled as a function <strong>of</strong> the BASE parameter. As evident<br />
from the figure, when scaling the workload the proportions <strong>of</strong> the different types <strong>of</strong> messages<br />
remain constant. This is expected since the relative weights <strong>of</strong> the various messaging styles used<br />
by the workload should not depend on the scaling factor.<br />
The sizes <strong>of</strong> the messages used in the various interactions have been chosen to reflect typical<br />
message sizes in real-life MOM applications. Pub/sub messages are generally much smaller<br />
than P2P messages due to the decoupled nature <strong>of</strong> the delivery mechanism. For every type<br />
<strong>of</strong> message, SPECjms2007 generates messages with sizes chosen from a discrete distribution<br />
with three possible values as shown in Table 5.5. There are two exceptions, the priceUpdate<br />
message used in Interaction 4 <strong>and</strong> the statInfoSM message used in Interaction 5. The former<br />
has a fixed size, while the latter has size between 4.7 <strong>and</strong> 24.78 KBytes with an average <strong>of</strong><br />
5.27 KBytes. Since statInfoSM messages contain sales statistics, their size is determined by<br />
the rate at which items are sold in supermarkets which depends on the number <strong>of</strong> customers<br />
visiting a supermarket per day <strong>and</strong> the average number <strong>of</strong> items sold per customer.<br />
P2P transactional messages tend to be medium in size <strong>and</strong> P2P non-transactional range from<br />
very small (≤ 1KBytes) to big (50 KBytes). Large messages are very rarely used in practice.<br />
We decided to model three different messages sizes per message type based on a certain<br />
probability:<br />
• Small messages: up to 2 KBytes, probability 95 %<br />
• Medium messages: up to 10 KBytes, probability <strong>of</strong> 4 %<br />
• Large messages: up to 55 KBytes, probability <strong>of</strong> 1 %<br />
Vertical Topology<br />
The goal <strong>of</strong> the vertical topology is to exercise the ability <strong>of</strong> the system to h<strong>and</strong>le increasing<br />
message traffic through a fixed set <strong>of</strong> destinations. Therefore, a fixed set <strong>of</strong> physical locations<br />
is used <strong>and</strong> the workload is scaled by increasing the rate at which interactions are executed.<br />
Similar to the horizontal case, a single parameter BASE is used as a scaling factor. The following<br />
rules are enforced:<br />
1. |Ψ SM | = 10<br />
2. |Ψ DC | = 2<br />
3. |Ψ SP | = 5